Determination of binding affinities using hyperpolarized NMR with simultaneous 4-channel detection

Abstract

Dissolution dynamic nuclear polarization (D-DNP) is a powerful technique to improve NMR sensitivity by a factor of thousands. Combining D-DNP with NMR-based screening enables to mitigate solubility or availability problems of ligands and target proteins in drug discovery as it can lower the concentration requirements into the sub-micromolar range. One of the challenges that D-DNP assisted NMR screening methods face for broad application, however, is a reduced throughput due to additional procedures and time required to create hyperpolarization. These requirements result in a delay of several tens of minutes in-between each NMR measurement. To solve this problem, we have developed a simultaneous 4-channel detection method for hyperpolarized ¹⁹F NMR, which can increase throughput fourfold by utilizing a purpose-built multiplexed NMR spectrometer and probe. With this system, the concentration-dependent binding interactions were observed for benzamidine and benzylamine with the serine protease trypsin. A T₂ relaxation measurement of a hyperpolarized reporter ligand (TFBC; CF₃C₆H₄CNHNH₂), which competes for the same binding site on trypsin with the other ligands, was used. The hyperpolarized TFBC was mixed with trypsin and the ligand of interest, and injected into four flow cells inside the NMR probe. Across the set of four channels, a concentration gradient was created. From the simultaneously acquired relaxation datasets, it was possible to determine the dissociation constant (K_D) of benzamidine and benzylamine without the requirement for individually optimizing experimental conditions for different affinities. A simulation showed that this 4-channel detection method applied to D-DNP NMR extends the screenable K_D range to up to three orders of magnitude in a single experiment.

Keywords: Dynamic nuclear polarization; Flow NMR; High throughput NMR; Protein-ligand interaction.

Figure 1.

(a) Flow path designed for simultaneous injection of four hyperpolarized samples into NMR flow cells. Non-HP reagents are independently admixed to each channel. t1: 152 cm, t2: 15 cm, t3 and t4: 5 cm, t5: 12 cm, t6: 20 cm, t7: 65 cm, and t8: 15 cm (b) Schematic of pulse transmission and signal reception for simultaneous NMR measurements with multiple channels. A CPMG pulse sequence employed for R₂ relaxation measurement is shown.

Figure 2.

(a) First 50 spin-echoes acquired from the four simultaneous measurements of concentration-dependent relaxation rates of hyperpolarized reporter ligand (TFBC) in the presence of trypsin and the competing ligand, benzamidine. (b) Final concentrations determined for TFBC, trypsin, and benzamidine in the measurements shown in (a). (c)-(f) Contour plots of chemical shift resolved CPMG spectra obtained from Ch1 – Ch4 (from left to right). First 625 spectra (0 ~ 5 s) are shown. The first spectrum from each dataset is shown on top of the contour plot. The integrated signals are plotted with a fitted curve (green) next to the contour plot.

Figure 3.

R₂ plots obtained from the relaxation measurements for benzamidine (a) and benzylamine (b). The curves represent the calculated relaxation rates using the fitted K_D of 21.6 μM and 205 μM for benzamidine and benzylamine, respectively, in combination with the average concentrations of trypsin and TFBC from the four channels. The error bars represent the 95% confidence intervals associated with the measured R_2,obs (along the y-axis) and 5% error of the measured concentrations for the competing ligands (along the x-axis)

Figure 4.

(a) Random sets of simulated R_2,obs values generated for each case of the competing ligand with K_D = 10⁻², 10⁻¹, 10⁰, 10¹, 10², 10³, and 10⁴ μM. For the simulations, the concentrations of reporter ligand, trypsin, and competing ligand were 22.3 μM, 0.62 μM, and [4.8, 48, 240, 480] μM; R_2,f = 0.66 s⁻¹, R_2,b = 841 s⁻¹, reporter ligand K_D = 142 μM. The curves represent the best fits to simulated data points. (b) Simulation results. Median values of the K_D distributions resulting from 10⁴ simulated datasets (‘Fitted K_D’) are plotted versus the corresponding values of true K_D. The error bars indicate 95% confidence intervals of the simulated K_D distributions. The lower end of the error bar for K_D ≤ 10⁻¹ μM extends to K_D = 0, which could not be drawn in the plot.